Lei Xiong

It seems that your environment settings are incompatible with sklearn. You can setup a new python env via conda, then install SCALE conda create -n SCALE python=3.6 pytorch conda activate...

Hi, thanks for your insterest in SCALE. 1. gamma=q(c|x)=p(c|z), gamma is an inference function that inferences the cluster (c) from the original data (x). However, the inference for c in...

SCALE is designed for sparse scATAC-seq datasets, of course it can run scRNA-seq data if you add option "--log_transform" in your command. But I highly recommend you use SCALEX (https://github.com/jsxlei/SCALEX),...

I think the reason is due to the latent dimention you choose is quite large. Just use the default parameter plus the --log_transform. It should work, at the same time,...

Why the recon_loss is 0? This does not make any sense at all. I think you should check your data first. SCALE takes count matrix without any normalization as input....

I just saw the max count can be 540254, and the smallest is 2 or 0 in your data. The numerical difference is big. You’d better do the log1p transformation...

Hi, the cluster assignments now are within the adata.h5ad adata file, which can be read by scanpy e.g. adata = scanpy.read('adata.h5ad'), and selected by adata.obs['leiden'].

You can get the tsne with two options, 1. run SCALE from scratch with --embed tSNE, then you can get a tsne.png and tsne embedding in adata.obsm. 2. run with...

Currently SCALE has disabled the kmeans cluster and adopt Leiden clustering, which will cause cluster number not same as wanted. You can do the Kmeans with the latent of SCALE...

It is should not be the row number of the peak file. SCALE read the peak file (mtx format) with pd.read_csv(filename, sep='\t', header=None).iloc[:, -1].values, it may be your input peak...